How AEO Helps Patients Find the Right Surgeon When They Cannot Judge for Themselves

The Question Patients Do Not Know to Ask

A patient choosing a surgeon cannot measure the variable that most predicts the outcome, so the decision runs on proxies instead. That sentence describes an ordinary afternoon in a surgical clinic. Someone sits down, hears a plan for an operation, then agrees to it while missing the one fact that would have told them whether this was the right room to be sitting in.

The instruments built to inform that decision fail a measurable standard. A study of 30,957 surgical patients treated by 2,921 surgeons across 65 United States hospitals found that from 36.5 percent to 57.7 percent reported inadequate shared decision-making, depending on which validated instrument was applied. Inadequate shared decision-making was significantly associated with postoperative complications.

The paperwork does no better. A systematic review of 26 studies covering 13,940 informed consent forms found 76.3 percent had poor readability, with 84 percent of the English-language forms rated difficult to read against an adequacy threshold of eighth grade or below. The document that legally records a patient's understanding is written above the level at which most people read.

None of this means patients are incurious. They ask what they have been given language for: how long the recovery takes, whether the scar shows, when they can drive again. Those are answerable questions with answers a practice publishes readily. The question that best predicts how the operation goes is the one nobody hands them the words for.

Set the two right-hand columns beside each other and the gap stops being subtle. What a patient can retrieve describes availability, credentials, then the texture of past appointments. What the research associates with survival or reoperation sits in a different column entirely, held inside the practice, rarely written down anywhere a person or a search engine can reach it.

One question carries more of that weight than any other, phrased in eight ordinary words. How many of these do you do a year. Almost no patient asks it, because nobody has told them it matters. Almost no practice answers it, because no page on the site was ever written to. Closing that gap honestly is the whole of what Answer Engine Optimization is for here.

What Actually Predicts How a Patient Does

The link between procedure-level experience and outcome is not folklore, nor is it confined to rare operations performed at a handful of centres. It has been measured, repeatedly, on the operations that fill ordinary theatre lists: hernia repair, hip replacement, cataract surgery, hysterectomy. These are the procedures a general population actually undergoes.

A Swedish nationwide register study of 20,656 elective groin hernia repairs across 75 surgical units tracked reoperation for recurrence against how many repairs each surgeon performed a year. Under 12 a year, 5.3 percent. At 12 to 50, 3.8 percent. At 50 to 150, 3.5 percent. Above 150, 2.9 percent. The adjusted hazard ratio for the lowest band was 1.87, on a 95 percent confidence interval of 1.31 to 2.67.

Hip replacement shows the same gradient. Grouping surgeons by annual volume, high at 150 or more, intermediate at 30 to 149, low under 30, three-month prosthetic joint infection ran 0.5 percent, 0.8 percent, then 1.0 percent. Adjusted odds of infection came out at 1.5 for intermediate volume, 1.87 for low. All-cause revision hazard ratios followed at 1.1 then 1.3.

Cataract surgery, among the most commonly performed operations in any health system, carries the same signal. Across 961,208 operations at 136 centres by 3,198 surgeons, posterior capsule rupture occurred in 1.01 percent overall, splitting to 0.77 percent for consultants against 1.93 percent for more experienced trainees, an odds ratio of 2.484.

Gynaecology supplies the cleanest subspecialty comparison. Among 1,631 benign minimally invasive hysterectomies, 52.4 percent performed by generalists against 47.6 percent by subspecialists, the adjusted odds of a Clavien-Dindo Grade III complication were 0.39 with a subspecialist rather than a high-volume generalist, on a 95 percent interval of 0.25 to 0.62.

Two honest limits belong beside those figures. Volume is a strong correlate across populations, never a guarantee for one person on one morning; a high-volume surgeon can have a bad day, a lower-volume surgeon can be excellent. The direction of causation is also argued, because practice may build skill, or referral patterns may route harder cases toward the busiest hands.

What survives both caveats is still enough to matter. Across every population studied, the band a surgeon sits in moves the probability of the thing the patient is most afraid of. That makes procedure-level volume the most decision-relevant number in the encounter, which is precisely why its absence from every retrievable source deserves a close look.

The Signals Patients Can See Do Not Measure That

The question then becomes whether any signal a patient can actually retrieve tracks that band. The largest test of it used 2,690,315 Medicare patients treated by 57,008 surgeons across 14 major operations, comparing publicly available reputation signals against 30-day mortality. The result deserves to be read twice.

Patient-initiated online star ratings were not consistently associated with 30-day mortality. Peer-nominated Top Doctor status was, at an adjusted risk difference of -0.14 percent, on a 95 percent confidence interval of -0.19 to -0.09. The signal a patient can look up in ten seconds carried no consistent relationship to survival. The signal built by other surgeons did.

Hospital star ratings behave the same way one level up. Across 1,898,829 Medicare patients at 3,240 hospitals, risk and reliability adjusted 30-day mortality ran 6.80 percent at one-star hospitals against 4.93 percent at five-star. The averages separate cleanly, which is what makes the next finding in the same paper so awkward for anyone using a rating to choose.

There was as much variation in mortality within a single star rating as there was across all ratings. Five-star hospitals ranged from 2.4 percent to 9.1 percent. A patient who picks a five-star hospital has not picked a low-mortality hospital. They have drawn a ticket from a bucket that happens to have a slightly better average printed on the side.

If ratings do not track mortality, it is fair to ask what they do track. An analysis of 1,833 five-star reviews of orthopaedic foot and ankle surgeons coded 3,215 individual comments. Good outcomes appeared in 29.2 percent. Bedside manner or patient experience appeared in 28.6 percent. A clear plan in 11.1 percent, a correct diagnosis in 5.4 percent, well-controlled pain in 4.5 percent.

Read the top two rows together. A five-star review is largely a report on how the encounter felt, written by someone with no independent way to verify the technical result being praised. That is a real thing to know about a surgeon, worth having. It is simply not the thing the volume literature measures, so it cannot stand in for it.

There is a sharper problem underneath. Among 231 shoulder surgeons, those holding a LinkedIn profile had higher mean Healthgrades ratings, 4.4 against 4.1 at P equals .02. Those with an active Instagram profile had higher mean Google Reviews ratings, 4.8 against 4.5 at P equals .002. The rating moved with the marketing footprint.

Nobody should read that as proof the ratings are bought. Surgeons who invest in a public profile may equally invest in the parts of practice that patients rate well. The point is narrower, then harder to dismiss: a number patients treat as a quality measure covaries with whether the surgeon keeps social accounts.

Then there is the thinness of the evidence itself. Of 201,154 United States surgeons, 78.86 percent carried at least one rating on a patient-initiated platform, at a mean of 4.12 out of 5. The median number of ratings per surgeon was ten, on an interquartile range of 1 to 32. Only 11.52 percent held peer recognition as a Top Doctor.

Ten reviews is not a sample. It is an anecdote with a decimal point attached, compressed onto a scale so narrow that almost every surgeon lands between 4 and 5. Taken as a whole, the retrievable proxy layer measures how the encounter felt, then how visible the surgeon has chosen to be. It does not measure how often this surgeon performs this operation.

None of those signals is worthless, so none of them should be discarded. Each is simply measuring something other than the variable that decides the outcome, which leaves the decisive fact unpublished by anybody at all. A fact nobody else has stated anywhere is the working definition of information a retrieval system cannot obtain from the consensus, which is also why it gets quoted.

What Patients Weigh Instead, and Why Referral Carries the Load

Ask patients how they chose, then the picture holds steady across samples. Among 538 hip and knee arthroplasty respondents, 50.2 percent came by physician referral, 27.7 percent through family or friends, 24.5 percent by self-guided research. Among 808 arthroplasty patients, 44.9 percent found their surgeon by referral, 18.7 percent by word of mouth, 8.4 percent citing online advertisements.

What those patients said mattered is equally consistent. On a one to five scale, the 538 respondents rated board certification 4.72, in-network insurance 4.66, fellowship training 4.50. Those are credential facts plus access facts. Every one of them is checkable in a minute, none varies much between the surgeons a patient is realistically choosing between.

Trust arrives early, then it arrives whole. In the 808-patient sample, trust in the chosen surgeon was rated at a median of 10 out of 10. Read alongside the referral share, that describes a decision largely made before the consultation begins, then confirmed rather than tested inside it. The referral carries almost the entire evidentiary load.

One orthopaedic reconstruction clinic asked 50 new patients to rate selection factors on that same one to five scale. Institutional reputation came first at 4.1, years in practice 3.8, insurance network 3.7, a primary care recommendation 3.7, online reviews 3.0, being easily found on the internet 2.6, social media presence 2.1. Ninety percent had arrived with a referral in hand.

Note first what a panel of fifty patients at one clinic cannot be asked to bear: it is one practice, never a national measure. Note then what is missing from the list altogether. Procedure-specific surgeon volume was not among the factors put in front of those patients. It was not rated low. It was never offered as something to rate.

That absence is the finding. Patients rank years in practice at 3.8 alongside institutional reputation at 4.1, which are both attempts to reach the same underlying idea: has this person done enough of this to be good at it. Years plus reputation are the proxies available. The measured variable was not on the menu, so a proxy took its seat.

Referral carries the load because it is the only channel in the system that has ever seen the number. A referring physician knows, roughly, who in town does a lot of these. That knowledge is real, it is unpublished, then it is unavailable to any patient without a well-connected doctor or a well-connected friend.

Which makes the inequity plain. The best information in the market travels by relationship, so a patient's odds of reaching a high-volume surgeon partly reflect the quality of their social network. Publishing the number is the only mechanism that puts it in front of the patient who has nobody to ask.

The Right Surgeon Includes the Right Operation

Choosing well is not only a question about who holds the instrument. It is also a question about whether the operation should happen at all, in that form, on that timeline. A patient cannot audit that judgement, because the person proposing the operation is usually the only expert in the room, and no second reading exists to compare it against.

Second opinions are rarest where the stakes are highest. Among 121,131 Medicare patients diagnosed with pancreatic cancer, only 4.2 percent, 5,108 people, obtained a surgical second opinion. Among the 10,949 who went on to have a pancreatectomy, 1,958 had sought one, which is 17.9 percent. On the most consequential operation in general surgery, most patients heard a single view.

When a second reading does happen, it changes things. Baseline rectal cancer MRI studies sent for subspecialist second-opinion review were discordant with the original report in 53.8 percent of cases, 248 of 461. Of those, 140 were major discordances, which is 56.5 percent. Two independent surgeons changed the recommended management in 19 percent then 44 percent of the major-discordant cases.

Those figures describe imaging reads rather than surgeon skill, so the inference has to stay tight. What they establish is that the staging judgement upstream of an operation is genuinely contestable, at a rate high enough to matter to a patient deciding whether to accept the plan in front of them without asking anyone else to look.

This is where a practice can answer a question patients have no standing to ask. Publishing indication criteria, plainly, states which presentations the practice operates on, which it manages without surgery, then which it refers elsewhere. A patient reading that page can locate their own situation in the criteria before the consultation ever begins.

The line carrying the most weight is the one naming what the practice does not do. A page stating which cases are sent onward, with the reason attached, is a page a patient can actually use. It is also the kind of page an assistant can quote, which is why a substantive page earns citations a service page cannot.

None of this replaces a second opinion. It lowers the cost of asking for one, by handing a patient a written standard to hold the recommendation against. A practice confident in its own indications loses nothing by publishing them, then gains the only patients it wanted in the first place: the ones it is genuinely right for.

The DSF Right-Surgeon Question Loop

The DSF Right-Surgeon Question Loop runs surgical practice publishing through six ordered stages plus a return leg that decides what gets published next. The order matters, because each stage consumes the output of the one before it. The governing rule is a single sentence: publish the number that answers the question, or publish nothing at all.

Stage one is the Asked Question, recorded in the words a patient actually uses before consenting, for one specific procedure. Not a keyword. Take hip replacement as the worked example. The sentence is close to this: how many hip replacements do you do a year, does that change the chance of infection. It gets written down verbatim, in that phrasing.

Stage two is the Deciding Variable, the one measurable thing that genuinely answers that question at procedure level. For the hip question it is the surgeon's annual primary hip arthroplasty count, because the volume bands in the literature are drawn on exactly that count. Where no measurable variable answers a question, the stage says so plainly rather than substituting a nearby one.

Stage three is the Held Record, which locates where the practice already holds that number. Theatre logs hold it. The stage fixes the denominator, primary procedures rather than all hip cases, fixes the period, the last full calendar year, then names the person who attests to it. A number with no named attester does not proceed past this point.

Stage four is the Verifiability Gate, the honesty valve the whole loop turns on. A number that could not survive an outside check never enters the loop. If the theatre log is incomplete, if the denominator is ambiguous, if the count depends on a generous reading of what counts as a primary case, the loop fails closed.

Failing closed means nothing is published. Not a range, not a rounded estimate, not a phrase like high volume standing in for a figure. The stage exists because the alternative, a flattering approximation, is the exact failure mode that makes surgical marketing dangerous. A practice unable to verify its own count has learned something worth knowing at this gate.

Stage five is the Published Answer: one dated, attributed, plain sentence on the procedure page carrying the number, its denominator, its period, then its source. The form reads like this. In the last full calendar year the practice performed a stated count of primary total hip replacements, taken from theatre records, attested by a named clinical lead.

That same sentence is mirrored in machine-readable form so a retrieval system can lift it without inference. Nothing about the mirror changes the claim. It restates the published fact in a structure a parser can read, which is the whole of its job, so it never carries a number the visible page does not.

Stage six is the Retrieved Answer, the only stage a practice does not control. An assistant asked how many of these a given surgeon performs answers with that sentence, then names the practice as its source. If it cannot, the sentence was either not published plainly enough or not published at all, which is a diagnosis rather than a mystery.

The return leg is the Unanswered Residue. Every outcome-deciding question the engines still answer without naming the practice feeds back into stage one as the next publishing list. Complication rates. Revision rates. Indication criteria. Length of stay. The loop is not a campaign with an end date; it is a standing cadence with a queue attached.

The loop is deliberately unkind to its own output, because a good number of the candidate figures die at stage four rather than reaching a page. That attrition is the feature being paid for. Digital Strategy Force Runs the Loop With the Clinical Team, sets the denominators before a word is written, then publishes only what survives an outside check.

Why This Is Publishing, Not Markup

It is tempting to hear all of this as a technical problem with a technical fix: add the right markup, then get quoted. Google's own documentation closes that door explicitly. As of December 2025, its guidance on AI Overviews and AI Mode states the eligibility condition in a form worth reading as an instruction rather than a hint.

No additional technical requirements, no special AI text files, no special structured data.— Google Search Central, guidance on AI features, December 2025

The stated condition is that a page be indexed, eligible to appear in Search, then eligible to be shown with a snippet. That is the whole gate. There is no file to add, no property to declare, no configuration that lifts a page into an answer it has not earned. Which leaves exactly one lever: what the page actually says.

The scale of the surface explains the urgency. Google reports that people ask more than a billion health questions every day on Search. A meaningful share of those are somebody deciding whether to have an operation, or deciding whether the surgeon proposing it is the right one. Those questions are being answered today, by something.

Increasingly they are answered by an assistant. KFF tracking polling found 29 percent of United States adults use AI tools or chatbots for health information or advice at least monthly, up from roughly 17 percent in June 2024. In the same polling, 44 percent said they were not too confident or not at all confident they could tell whether such information was true.

Hold those two findings together. A large, growing share of patients are asking a system they do not trust themselves to audit. Whatever that system says about a practice becomes, functionally, the practice's answer, whether the practice wrote it or not. Silence is not neutrality in that arrangement; it is delegation to whoever did write something.

How a health page earns that trust in the first place is covered elsewhere, in the standards that make health content citable. What matters here is narrower: the substance being trusted has to exist before any of those standards can operate on it. A well-formed page with nothing inside it stays uncited, correctly.

Structured data still has a job, a narrow one. Schema.org publishes a Physician type, defined as an individual physician or a physician's office treated as a medical organisation, carrying properties for medical specialty, available service, then hospital affiliation. It gives a parser somewhere to put facts the page already states.

Note the direction of travel there. The type offers a home for an available service, so a practice holding a verified procedure-level count has something real to put in it. A practice holding nothing has an empty container with correct syntax. Markup describes the fact, so it cannot create it, which is why the publishing work comes first.

The same logic governs the credentials sitting around the number, where consistency across every place a surgeon is listed does more work than any single well-written page. That is the terrain of expertise signals built at the entity level, a discipline every regulated profession shares, because a name that resolves cleanly is a name a system can attach a fact to.

Consistency is not a given, either. Across 634,914 physicians in PECOS, with 449,282 listed in two or more insurer directories, address consistency ran from 16.5 percent to 27.9 percent, phone consistency from 16.0 percent to 27.4 percent, specialty consistency from 64.2 percent to 68.0 percent. The directory layer cannot agree on where a surgeon works.

So there is no markup trick available here, which is exactly why publishable fact is the only lever that remains. The practices quoted on these questions will be the ones holding a number nobody else in the market is able to state, which is the same logic behind building data assets a model cannot route around. Scarcity of the fact is the whole advantage.

The Honesty Floor, and What It Costs to Hold

Every argument above cuts both ways, which has to be said plainly rather than buried. A method that makes a good practice legible also makes an overstating practice louder. So the floor has to be set before the publishing starts, then held when holding it is inconvenient. The evidence on what happens without a floor is not ambiguous.

Seventy-one patient-facing websites on patient-specific knee arthroplasty were scored for quality. Academic sites averaged 25.4 on the QUEST instrument against 9.8 for non-academic ones. Disadvantages of the procedure appeared on 69.1 percent of academic sites against 12.5 percent of non-academic. Peer-reviewed references appeared on 81.8 percent against 12.5 percent.

The sharpest line in that study is the one about accuracy. Inaccurate claims appeared on 31.3 percent of non-academic sites, then were absent from the academic ones entirely. Readability was uniformly poor across both groups, at a mean SMOG grade of 13.5, so nobody in that sample was writing for the patient who most needed to understand the page.

Read those columns as a mirror rather than a scold. The difference between the two groups is not budget, nor design talent. It is whether the page discloses what the procedure costs a patient in risk, cites something checkable, then declines to claim more than the evidence supports. Each of those is a choice available to any practice on any afternoon.

A false claim does not sit quietly on a page any more. HealthBench, built from 5,000 multi-turn health conversations scored against 48,562 physician-written rubric criteria by 262 physicians across 26 specialties, found that on its hardest set no evaluated model scored above 32 percent. These systems are not dependable correctors of health claims.

Set that beside the 44 percent of adults who doubt their own ability to spot a false answer, then the propagation path is clear. An overstated claim on a procedure page is not caught by the model, is not caught by the reader, then arrives at a frightened person as an answer with the practice's name attached to it. The claim travels; the correction does not.

So the rule is stated without hedging. A practice must not publish a number it cannot defend to an auditor, a regulator, or an expert witness. A low-volume practice must not use any of this to appear expert. If the count is low, the honest published sentence is the low count, or there is no sentence at all.

Overstating surgical competence is two failures at once. It is a compliance exposure, because a claim about clinical outcomes is a regulated claim in every jurisdiction that matters. It is also a patient-harm risk, because a patient who chooses on a false number has been moved toward the operation by the very thing that should have warned them off it.

Which brings the decision back to a desk. Somebody has to own the denominator. That means a named person who decides what counts as a primary case, who reconciles the theatre log, then who signs the sentence before it goes live. In most practices no such role exists, which is the plainest reason no such sentence exists either.

The cadence has a cost that deserves an honest quote. Counts change every year, so every published sentence carries a review date. Indication criteria move as the evidence moves. A page true in one year becomes a liability in the next if nobody is accountable for revisiting it. That recurring obligation, rather than the writing, is the real commitment being made.

That is the whole proposition, stated at its least flattering. There is no shortcut, the honest numbers are sometimes worse than the marketing would prefer, then the work never finishes. What a practice gets in return is narrow but real. When a frightened person asks the question that most predicts their outcome, the answer coming back is true, sourced, then attached to a name that can be held to it.

FAQ — Finding the Right Surgeon

Next Steps — Finding the Right Surgeon

• ▶ Write down the questions patients actually ask before consenting, procedure by procedure, in their words. That list, not a keyword report, is the publishing agenda for the year ahead.
• ▶ Name the owner of the denominator before anything is drafted. One person decides what counts as a primary case, reconciles the theatre log, then signs each sentence that goes live.
• ▶ Run every candidate number through the Verifiability Gate first. If it would not survive an outside audit unchanged, the loop fails closed, so nothing about it reaches the site.
• ▶ Publish indication criteria alongside the counts, including the presentations the practice refers elsewhere. The line naming what a practice declines is the line patients cannot obtain anywhere else.
• ▶ Test the retrieved answer quarterly by asking an assistant the patient's own question, then treat every unanswered question as the next item in the queue.

A surgical practice does not need louder marketing. It needs the facts that decide outcomes written down where a frightened person can find them, held to a standard that survives scrutiny from a regulator as readily as from a patient. Digital Strategy Force builds that record with the clinical team, publishes only what clears the gate, then keeps it current as the counts change. Speak With the Answer Engine Optimization Team.